dah / README.md
osidenna's picture
add number 2 for latin codes
1667c80 verified

Dataset Card for DAH (DAtaset Hassaniya)

DAH is a bilingual dataset created to support translation between Hassaniya dialect and English, with both Arabic script and Arabizi (Latin-transliterated) Hassaniya included.


Dataset Details

  • Name: dah
  • Languages: Hassaniya Arabic (in both Arabic and Latin forms), English
  • Creators:
    • Ahlam Abdelkader
    • Emani Babe
    • Oumoukelthoum Sidenna
  • Adapted from: Tatoeba Project parallel sentences
  • Reviewed by: Founders of the Hassan-IA community

Standardization

To ensure consistent batches and model outputs, a standard mapping was agreed upon for Hassaniya → Latin transliteration.
The Latin representation follows the conventions widely used in Arabizi:

Arabic Letter Latin Equivalent
ء 2
ع 3
ذ 4
خ 5
ط 6
ح 7
ظ 8
ق 9
غ gh
ض dh

All sentences were normalized according to these rules and reviewed by at least one additional team member for consistency.

Note:
In some examples, both “ou” and “u” are used interchangeably to reflect the Arabic و sound.
This variation follows current informal writing conventions in online Hassaniya communication — particularly among younger speakers and social media users, where both spellings are common and accepted.


Dataset Structure

Each entry includes:

  • english: The English sentence
  • hassaniya-ar: Hassaniya Arabic in Arabic script
  • hassaniya-en: Hassaniya in Latin (Arabizi) transliteration